Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Received: November 30, 2017 49
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
Short-Term Stock Market Fuzzy Trading System with Fuzzy Capital
Management
Ahmed Tealab1* Hesham Hefny1 Amr Badr1
1Computer Science Department, Institute of Statistical Studies and Research, Cairo University, Giza, Egypt
* Corresponding author’s Email: [email protected]
Abstract: Forecasting financial market indices became a necessary operation for investors’ decisions in order to get
the maximum return of investments. The complex behavior of the stock market requires development of forecasting
systems. This work proposes a short-term stock fuzzy decision system using a novel trading strategy based on
mixture of technical indicators. Fuzzy logic is applied for both trading rules definition and portfolio management.
The selected stock market technical indicators for designing trading rules consist of commonly used indicators and
new developed one. That is based on daily candlestick information produces short and long entry signals. The
proposed system is tested using the Athens Stock Exchange data. The results are compared to classical non-fuzzy
systems in addition to latest fuzzy approaches. The proposals performance produced less losses and better profits.
The results demonstrate that the fuzzy logic is promising in portfolio management with steady upward profit and low
losses. However, that deserves more study.
Keywords: Prediction, Fuzzy logic, Stock market, Short-term time series.
1. Introduction
Stock market has tight relation with countries
economic. Finding the best time to buy or sell
indices is the main challenge for experts and that
requires prediction of market moves. Stock market
indices behavior in general is quite noisy, nonlinear,
and chaotic by nature [1]. It is not in random moves
and can be classified as nonlinear time series with
strong relation to previous data [2]. Different factors
such as inflation reports, short term interest rate,
political events, traders’ expectations, and
environmental factors are interpreted in stock prices
moves causing market fluctuations which will make
forecasting stock prices a difficult task. However,
the evolution of technology and computer systems
have affected the field of stock market forecasting in
a way leads to develop financial forecasting and
decision systems [3]. The use of intelligent systems
such as neural networks [4], fuzzy logic [5] and
genetic algorithms [6] in financial forecasting has
several applications. Many technical indicators are
being used in stock market prediction producing
different results, their types include: Trend, Volume,
Oscillators and Volatility indicators. Technical
analysts and trading experts use one or more
technical indicator in their market analysis and
trading strategies.
This paper is organized as follows: Section 2
gives a background about related work. Technical
analysis, soft computing and decision support
systems are explained in Section 3. Section 4
describes Stock market prediction and money
management model. Section 5 presents the
experimental results. Finally, Section 6 concludes
our work with a scope for future work.
2. Background
2.1 Technical indicators and technical analysis
Stock investor has to risk purchasing
investments to get higher returns. Technical analysis
is exploited as stock prices trends forecasting
technique by studying historical indices’ data. Thus,
according to Efficient Market Hypothesis (EMH)
Received: November 30, 2017 50
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
studying specific technical indicators is enough to
forecast prices moves and trends. Experts use
combination of technical analysis techniques to
forecast market moves by studying market history.
They usually make subjective assessments to the
data before developing their trading strategy; it is a
difficult task and may have contradictory results [7].
Fuzzy models can address these issues.
Most of studies use common technical indicators
derived from Moving Average (MA) such as
Relative Strength Index (RSI) and Stochastic
Oscillator (SO). Successful stock market forecasting
strategies will be adapted by market and won’t be
successful any more. However, using the previous
studied trading rules may give different profitable
results for future periods and continues success is
not guaranteed [8]. Studies found that technical
trading strategies based MA rules won’t yield profit
with time [9].
Distinguishable characteristics of the indicators
determine their profitability according to the market
conditions. Furthermore, most of the common
indicators are calculated for the previous sessions
close price only, while there are other session data
can be considered such as open, high and low price
values. Therefore, it is obvious that there is a need
for new technical indicators choices that are
calculated based on full session data for better
trading decisions based information. Since each
technical indicator is developed for specific purpose,
so combining different indicators may negate each
other resulting in non-desired performance [10].
2.2 Decision support systems
Decision support systems (DSS) have promoted
firstly from simple model-oriented systems to
advanced multi-function entities [11]. There is a
need to find the suitable DSS framework for each
industry [12]. Thus, the financial industry has less
work to be done, in particular, in the stock market
filed. DSS were not advanced enough to fulfill
financial decision takers needs [13]. However, some
studies proposed intelligent decision systems based
on studying the historical and current market data
[14]. Before the exploration of soft computing, most
of studies were focusing on time series analysis such
as ARIMA, ARCH, and GARCH [15]. However,
developing various computational intelligence forms
such as neural network, fuzzy logic and genetic
algorithm to predict stock market prices is needed.
2.3 Soft computing
Soft computing (SC) is the use of inexact and
Figure 1. Fuzzy decision system
uncertain solutions for computationally hard tasks
where there is no known algorithm to achieve
tractability for low cost. Fuzzy logic, artificial
neural network, particle swarm optimization and
genetic algorithms amongst others are the principal
constituents of Soft Computing. Fuzzy logic is
defined by Zadeh [16] as a set of mathematical
principles based on degrees of membership rather
than classical binary logic that is used for
knowledge representation Hybrid neuro-fuzzy
model for stock market prediction [17], recursive
possibilistic fuzzy modeling [18] in addition to the
impact of sentiments from tweets as well as RSS
news feeds using neural network [19] are successful
applications in the real world. These studies resulted
in significant improve in stock market prediction but
without smart fuzzy management methodology.
Thus, it is apparent that a need of an intelligent
system that can precisely and reliably perform
prediction with portfolio management is important
[20].
3. Proposed systems
The proposed system is identified through this
section to explain its intelligent components. Three
decision systems are considered, the first one has
classical strategy. The other two are fuzzy systems,
one of them has only fuzzy rules and the other has
fuzzy rules and fuzzy capital management. Each
system has one or more trading strategy. The trading
strategy used in the proposed systems based on three
indicators, two of them are known technical
indicators and novel developed one.
The process of the fuzzy systems starts by
fuzzifying the received input values them execute
set of predefined IF–THEN-type rules when meet
the reasoning conditions. The system major phases
are (Fig. 1):
A) System architecture and operations definitions:
In this phase the whole system architecture and
fuzzy system communication with the other
system components are defined.
Received: November 30, 2017 51
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
B) Fuzzification: The process of transforming
input values into fuzzy sets with assigned
names and membership functions.
C) STFIS rule Base: It is a set of different STFIS
rules, each has an “if - then” structure. “If” part
specifies values of technical indicators and the
“then” part specifies a rating given these values.
D) Defuzzification: In this phase the fuzzy output
signal is transformed into a single number from
predefined range of values and it represents the
final system decision.
3.1 Technical indicators
The technical analysis concerns with the price
movements in the market. Technical analysts
attempt to use tools such as technical indicators,
oscillators and charts to recognize patterns which
can estimate future movement. In this study, a set of
three technical indicators are used, the first one is
the Stochastic Oscillator (SO). SO is momentum
indicator; its value oscillates between 0 and 100
which represents the price of a security relative to
the high/low range over a specific period of time. Its
lines divergences and centerline crossovers can be
used as indicator for overbought or oversold statuses.
The calculations that draw stochastic oscillator lines
are as following:
K =(𝐶𝑡 − 𝐿𝑛)
(𝐻𝑛 − 𝐿𝑛)× 100 (1)
D𝑓𝑎𝑠𝑡 = 𝑆𝑀𝐴(3, K) (2)
D𝑠𝑙𝑜𝑤 = 𝑆𝑀𝐴(3, D) (3)
where 𝐶𝑡 is current price, 𝐿𝑛 and 𝐻𝑛 are the lowest
and the highest close prices in last 𝑛 sessions
respectively. 𝑆𝑀𝐴(𝑛, x) is the simple moving
average of variable x for 𝑛 sessions.
The recommended number of sessions by
experts is 𝑛 = 14 which is the value used in this
research. SO values above 80 reflect overbought
conditions; on the opposite side; values under 20 are
interpreted as indication of oversold conditions. The
central line with value 50 represents neutral market
momentum. In terms of market analysis and trading
signals, SO that moves under 20 up to above 20
level is considered as bullish signal, while SO
moves below 80 level coming from above 80 is
viewed as bearish signal.
The second used indicator in this study is ADX
(Average Directional Movement Index). It measures
the strength of a trend of a financial time series. Its
value based on DI+ and DI− which are the positive
and negative directional indicators respectively. The
indicator value is calculated from the following
formulas:
𝐴𝐷𝑋 =𝐷𝐼+−𝐷𝐼−
𝐷𝐼++𝐷𝐼− (4)
𝐷𝐼+ =Smoothed 𝐷𝑀+
𝑇𝑅 (5)
𝐷𝐼− =Smoothed 𝐷𝑀−
𝑇𝑅 (6)
𝐷𝑀+ = Ht – Ht-1 or 0 if result < 0 (7)
𝐷𝑀− = - (Lt – Lt-1) or 0 if result < 0 (8)
where 𝐷𝑀+ is Plus Directional Movement and 𝐷𝑀−
is the Minus Directional Movement, they determine
the strength of bullish and bearish moves
respectively, TR is the True Range, Ht, Lt current
high and low where Ht-1, Lt-1 are previous high and
low respectively. TR is largest absolute value of
difference between highest and lowest value in n
periods. The recommended number of sessions by
experts is 14 which is the value used in this research.
The ADX generates a signal that quantifies the
strength of an underway trend. Its value ranges
between 0 and 100; ADX values less than 20 refer to
trend weakness while values above 40 indicate trend
strength as a common interpretation that is accepted
among technical analysts.
The third indicator is the MyPivots (MP). MP is
the new proposed indicator to determine market
entry signals. MP is a customization of the
Camarilla Pivot Points (CPP) indicator which
commonly used by daily traders. The CPP lines are
expected to determine intraday support and
resistance levels; suggesting market entry and exit
signals. The previous trading day High, Low and
Close values are the input data for Camarilla
equations to compose the new intraday levels. This
method relies on the principle that the market price,
which is a type of time series, tending to revert back
to its mean value. Additionally, it is a type of trend
following indicator too. It is useful with high
sensitivity of MA based indicators which sometimes
results in some false market entries. However, CPP
is more precise for high volume and volatile stocks.
Simply, these data are used as input parameters for
the CPP equations to generate the new intraday
support and resistance levels as follows:
𝑅6 = (𝐻𝑑−1/𝐿𝑑−1) × 𝐶𝑑−1 𝑅5 = 𝑅4 + (𝑅4 − 𝑅3) × 1.168 𝑅4 = 𝐶𝑑−1 + (𝐻𝑑−1 − 𝐿𝑑−1) × 1.1/2 𝑅3 = 𝐶𝑑−1 + (𝐻𝑑−1 − 𝐿𝑑−1) × 1.1/4 𝑅2 = 𝐶𝑑−1 + (𝐻𝑑−1 − 𝐿𝑑−1) × 1.1/6
(9)
Received: November 30, 2017 52
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
where R and S refers to resistance and support levels
respectively; 𝐶𝑑−1, 𝐻𝑑−1 and 𝐿𝑑−1 are Close, High,
Low prices of the previous day respectively, while
PP is the mean value and called Pivot Point line.
The Camarilla based trading strategy is to issue
short position if the market reached up the area
between R3 and R4 aiming the market to reverse
down to PP. On the opposite side is to issue long
position if the market price reached down the area
between S3 and S4, expecting the market to revert
up to PP. Another trading way with Camarilla is to
follow the trend. It is concerned with the price
breakout up of R4 or down from S4 generating
bullish or bearish trend respectively. That will
suggest issuing long position for bullish market and
short position for bearish market trend.
The MP signal is based on two conditions come
from the change in the market strength and
determined by 𝐶𝑃𝑃 values as follows:
𝑀𝑃 =
{
𝐿𝑜𝑛𝑔 𝑖𝑓 𝑆3 ≥ 𝑂𝑡 > 𝑆4
𝑎𝑛𝑑 𝑃𝑃 ≤ (𝐻𝑑−2 + 𝐿𝑑−2)/2,
𝑆ℎ𝑜𝑟𝑡 𝑖𝑓 𝑅4 > 𝑂𝑡 ≥ 𝑅3
𝑎𝑛𝑑 𝑃𝑃 ≥ (𝐻𝑑−2 + 𝐿𝑑−2)/2,𝑁𝑜 𝐸𝑛𝑡𝑟𝑦 𝑖𝑓 𝑜𝑡ℎ𝑒𝑟
(10)
where 𝑂𝑡 is the open value in at time t. In case of
LONG and SHORT signals, their strength will be
estimated using the following formula:
strength =
𝑄 −3 × (𝑂𝑡 − 𝑂𝑡−1)
max_𝑑𝑖𝑓𝑓_𝑂𝑝𝑒𝑛 −
10 × 𝑃𝑃
max_𝑑𝑖𝑓𝑓_𝑃𝑃
−3 × (𝑅𝑎𝑛𝑔𝑒𝑡−1 − 𝑅𝑎𝑛𝑔𝑒𝑡−2)
max_𝑑𝑖𝑓𝑓_𝑅𝑎𝑛𝑔𝑒
(11)
where and max _𝑑𝑖𝑓𝑓_𝑂𝑝𝑒𝑛 , max _𝑑𝑖𝑓𝑓_𝑅𝑎𝑛𝑔𝑒 ,
max _𝑑𝑖𝑓𝑓_𝑃𝑃 are the difference between the
minimum and maximum values of all 𝑂𝑝𝑒𝑛, 𝑅𝑛𝑎𝑔𝑒
and 𝑃𝑃 values respectively in all sessions from 1 to t.
𝑄 is the strength factor which is determined
according the signal type as follows:
𝑄 = {
75 𝑖𝑓 𝑀𝑃 𝑖𝑠 𝐿𝑜𝑛𝑔 25 𝑖𝑓 𝑀𝑃 𝑖𝑠 𝑆ℎ𝑜𝑟𝑡 50 𝑖𝑓 𝑀𝑃 𝑖𝑠 𝑁𝑜 𝐸𝑛𝑡𝑟𝑦
(12)
The strength of long increases as the value of 𝑄 is
lower than 25 and while the higher the value of 𝑄
more than 75 the stronger the short signal.
3.2 Entry rules
Entering the market for long or short position is
based on the output signal of the decision system.
The output signal generated from the combination of
the trading indicators results in Bullish or Bearish
trend. In bullish market prices are expected to rise,
while in bearish market prices are expected to drop.
Therefore, the suggested positions type is
determined based on the type of the output signal;
long position for bullish market and short position
for bearish trend.
In this paper, the results of the decision system
modelling with fuzzy set rules and the classical
(crisp) methods are examined. The fuzzification of
the technical indicators gives the information to
generate trading rules. The crisp system based on
technical analysis only. Additionally, the
performance of the proposed systems are compared
to results obtained by the short-term fuzzy trading
system (STFS) proposed by K. Chourmouziadi [15].
3.2.1. Classical system trading rules
The classical system trading strategy is based on
set of technical indicators which have a range of
frequencies to compose the trading rules. The inputs
are positive integer values of the technical indicators
and the outputs are numeric values (-1: Bearish, 1:
Bullish or 0: No Entry). Specific combination of the
three antecedents indicators inputs mar represents
certain market trend signal. The classical system
output is signal for entering or existing market
signal without strength qualification. Bullish and
Bearish rules are shown in Table 1.
3.2.2. Fuzzy trading rules
The output signals of the decision systems are in
the form of linguistic terms such as Bullish or
Bearish in absence of any type of preciseness.
The imprecise output results in uncertainty of
traders’ decision which can be tolerated by using
fuzzy logic. The experts believe that the market
behavior is interpreted in the trader’s decision
system. However, that may keep producing signals
𝑅1 = 𝐶𝑑−1 + (𝐻𝑑−1 − 𝐿𝑑−1) × 1.1/12 𝑃𝑃 = (𝐻𝑑−1 + 𝐿𝑑−1 + 𝐶𝑑−1)/3 𝑆1 = 𝐶𝑑−1 − (𝐻𝑑−1 − 𝐿𝑑−1) × 1.1/12 𝑆2 = 𝐶𝑑−1 − (𝐻𝑑−1 − 𝐿𝑑−1) × 1.1/6 𝑆3 = 𝐶𝑑−1 − (𝐻𝑑−1 − 𝐿𝑑−1) × 1.1/4 𝑆5 = 𝑆4 − (𝑆3 − 𝑆4) × 1.168 𝑆6 = 𝐶𝑑−1 − (𝑅6 − 𝐶𝑑−1)
Received: November 30, 2017 53
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
Table 1. Crisp system rules if 30 <OS<85 & 25 <ADX<55 & S4≤Ot< S3 then 1
elseif 25 <OS<85 & 55 <ADX<100 & S4≤Ot< S3 then 1
elseif 0 <OS<25 & 25 <ADX<55 & S4≤Ot< S3 then 1
elseif 0 <OS<25 & 0 <ADX<25 & S4≤Ot< S3 then 1
elseif 0 <OS<25 & 55 <ADX<100 & S4≤Ot< S3 then 1
elseif 0 <OS<25 & 25 <ADX< 55 & S4≤Ot< S3 then 1
elseif 85<OS<100 & 25<ADX<55 & R3<Ot≤ R4 then -1
elseif 85<OS<100 & 0 <ADX<55 & R3<Ot≤ R4 then -1
elseif 85<OS<100 & 55 <ADX<100 & R3<Ot≤ R4 then -1
elseif 35 <OS< 85 & 25<ADX<55 & R3<Ot≤ R4 then -1
elseif 35 <OS< 85 & 55<ADX<100 & R3<Ot≤ R4 then -1
elseif 85<OS<100 & 25<ADX<55 & R3<Ot≤ R4 then -1
else 0
(a) (b)
(c) (d)
Figure. 2 Fuzzification: (a) MyPivots indicator, (b) SO indicator, (c) ADX indicator, and (d) output signal
based on imprecise inputs. ADX and SO indicators
values between 0 and 100. In MP lines, by
considering R6 is the top value and it is equal to 100
on the 0-100 scale. Thus, the other MP lines on the
0-100 can be found using the following formula:
nLine = 100 × Line/R6 (13)
where nLine is the new MP line’s value on the 0-
100 scale and Line is the real MP value.
Fig. 2 shows the fuzzification of the technical
indicators and output conditions of the proposed
system. The different input values of MP are
classified to three fuzzy sets: LONG (LN) for long
position, SHORT (SH) for short position and WAIT
(WT) for no position or no entry signals. The WAIT
value represents a neutral case for market status
between LONG and SHORT or extreme indicator
values within its range. However, the traders believe
that the market will reverse its position from these
extreme locations (Fig. 2 (a)). The assigned fuzzy
sets to each indicator are different from indicator to
another one; it depends on the indicator output
signal. The SO values vary between 0 and100 has
assigned LOW (LO), MEDIUM (ME) and HIGH
(HI) fuzzy sets for its outputs (Fig. 2 (b)). Similarly,
the ADX indicators low, medium and high output
values are assigned NOTREND (NO), ONTREND (ON) and RISKY (RIS) fuzzy sets respectively. The
high ADX values considered RISKY cause it is
almost come to the end of the trend (Fig. 2 (c)).
Received: November 30, 2017 54
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
Table 2. Fuzzy trading rules
OS ADX MP Bullish Bearish
HI ON LN WK WK
HI NO LN WK WK
HI RIS LN WK WK
ME ON LN VS WK
ME NO LN WK WK
ME RIS LN ST WK
LO ON LN VS WK
LO NO LN ST WK
LO RIS LN ST WK
HI ON SR WK VS
HI NO SR WK ST
HI RIS SR WK ST
ME ON SR WK VS
ME NO SR WK WK
ME RIS SR WK ST
LO ON SR WK WK
LO NO SR WK WK
LO RIS SR WK WK
HI ON WT WK ST
HI NO WT WK WK
HI RIS WT WK WK
ME ON WT WK WK
ME NO WT WK WK
ME RIS WT WK WK
L ON W ST WK
L NO W WK WK
L RIS W WK WK
WEAK (WK), STRONG (ST) and VERYSTRONG
(VS) are fuzzy sets with clear meanings; describe
the bullish and bearish outputs (Fig. 2 (d)). For the
generality and simplicity reasons, triangular
membership functions represented all the fuzzy sets.
Table 2 contains defined set of possible
indicators combinations that follow the behavior of
the trading experts in formulating their trading rules.
Consequently, the fuzzy sets of output signals
(bullish and bearish) strength are (WEAK,
STRONG and VERYSTRONG). That is interpreted
in the system entry action to the market with
different recommendation based on the strength
level. The system output signals’ strengths are
STRONG (ST), VERYSTRONG (VS) and WEAK
(WK) for both the Bullish and Bearish signals. The
used fuzzy system is Mamdani [17] with
maximum/minimum based operators. The centroid
method is used in the defuzzification phase.
3.3 Exit conditions
The unpredictability of the after trades market
moves might raise greed or fear emotional states.
Thus, a method that defines market exit conditions
and avoiding instant monitoring to market is
required for any trading system. The setup of stop
threshold conditions (stop-loss and stop-profit) is a
protective step for the established positions to limit
their maximum losses and revenues. Managing both
losses and profits will ensure prudential revenue
from unexpected extreme market moves or turns.
This step will be used to calculate the amount of
money to invest.
The volatility of the market at time of issuing a
position will determine the (stop-loss and stop-
profit) thresholds. However, the ATR (Average True
Range) is used as volatility estimator. ATR is found
for 𝑞 session which is equal to 14 as used with SO
and ADX indicators. The ATR is defined as follows:
𝑇𝑅 = 𝑀𝑎𝑥(𝑀𝑡 − 𝑚𝑡 , 𝑀𝑡 − 𝑐𝑡−1 , 𝑚𝑡
− 𝑐𝑡−1) (14)
where 𝑀𝑡 ,𝑚𝑡 represent the highest, lowest price
values in a session 𝑡 respectilvey, while 𝑐𝑡−1 is close
value of the previous session. In the proposed
system the exit conditions are calculated and
assigned at the position open time. For bullish
market case, the stop-loss is the position open price
minus twice the ATR value. To have balanced and
conservative strategy with Risk/Reward ration 1/1,
the stop-profit will also be the position open price
plus twice the ATR value and vice versa for bearish
market.
4. Money management
In the proposed system the output of the fuzzy
trading rules is the input of the fuzzy capital
management system. The fuzzy trading system is
responsible to decide the amount of money to invest
with a simple risk management strategy. Vince [21]
proposed the optimal-F that provides a method to
calculate the required amount of the investor’s
portfolio to achieve a geometric profits growth. Its
way to evaluate exactly the money to be used in
each trade is a capital percentage that is directly
proportional to the previous winning trades. A
customization has been done to the optimal-F that its
input is replaced by the fuzzy output generated by
the proposed intelligent system.
One approach here is the “fixed fractional
trading” strategy, where an investor wants to risk a
fixed percentage of his current capital for all trades
and investments. The past investments yields affect
the following capital available for investment. To
maximize the Terminal Wealth Relative (TWR) for
a specific trading period we have to calculate the
optimal-F to find required the fraction (f) for that.
Each trading operation 𝑖 has a return factor
𝐻𝑃𝑅𝑖(𝑓) which is a function of the operation result
𝑟𝑒𝑡𝑢𝑟𝑛𝑖 and worst trade ( 𝑑𝑟𝑎𝑤𝑑𝑜𝑤𝑛 ) in the
Received: November 30, 2017 55
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
specified trading period. The following equations
calculate the TWR:
𝑇𝑊𝑅(𝑓) = ∏𝐻𝑃𝑅𝑖(𝑓)
2
𝑖=1
(15)
𝐻𝑃𝑅𝑖(𝑓) = 1 +𝑓. (𝑟𝑒𝑡𝑢𝑟𝑛𝑖)
𝑑𝑟𝑎𝑤𝑑𝑜𝑤𝑛 (16)
Therefore, the relative relation between the possible
risk and the expected reward (Risk/Reward) for each
trade means that targeting high profitable trades will
require high optimal value of 𝑓 which is practically
rare. A practical value of 𝑓 can be considered as
percentage of optimal-F. In the proposed system,
partial-F is introduced as 12% of the found optimal-
F as the possible applied 𝑓.
Fig. 3 demonstrates the Risk/Reward relation
curve or the expected gain against the risk assumed.
In the Gaussian curve it is clear that the highest
point of the curve which represents the maximum
expected profit (reward) is at the optimal-F.
Therewith, it is possible to observe that with less
risk, less profit is expected. Partial–F is located on
the left side of the curve at 12% of the optimal-F. The fuzzy trading system determines the range
amount of capital available for investment; it will be
estimated as:
𝑐𝑢𝑠𝑡𝑜𝑚𝑓= 𝑝𝑎𝑟𝑡𝑖𝑎𝑙𝑓
+(𝑠𝑖𝑔𝑛𝑎𝑙 − min _𝑠𝑖𝑔𝑛𝑎𝑙)(𝑜𝑝𝑡𝑖𝑚𝑎𝑙𝑓 − 𝑝𝑎𝑟𝑡𝑖𝑎𝑙𝑓)
𝑠𝑔𝑖𝑛𝑎𝑙𝑟 × 𝑜𝑝𝑡𝑖𝑚𝑎𝑙𝑓
(17)
where signal is the system output (Bullish or
Bearish); min _sginal is the minimum signal value
to trigger a trade (its value is 15); optimalf and
partialf are optimal-F and partial-F respectively;
Figure 3. optimal-F, obtained-F and partial-F
demonstration
sginalr is difference between minimum and
maximum of signals values. After obtaining the
optimal-F, the next step is to calculate the required
number of shares using the following formula:
#𝑠ℎ𝑎𝑟𝑒𝑠 = 𝑝𝑎𝑟𝑡𝑖𝑎𝑙𝑓 +𝑒𝑞𝑢𝑖𝑡𝑦
𝑠𝑡𝑜𝑝𝑙𝑜𝑠𝑠 (18)
where 𝑒𝑞𝑢𝑖𝑡𝑦 is the amount of money available for
investment and 𝑠𝑡𝑜𝑝𝑙𝑜𝑠𝑠 is the market exit
condition value. The above formula (18) output is
estimation to the number of shares required for
purchase without consideration of the share price. In
case that the available money for investment isn’t
enough to cover the required number of shares, the
maximum available number is used. But in the
classical system it uses trading rules with all
available capital for investment in all assigned
positions.
5. Experimental setup
In this research the Metatrader Trader software
(MT4) is used to implement the three trading
systems and to compare their performance for the
same database. To test these systems well, the
criteria of selecting the validation market indices
should include diversity of trends through the
validation period.
The results obtained by K. Chourmouziadis [14]
fuzzy system are used as our benchmark; they
showed better performance than the B&H strategy.
Chourmouziadis [14] proposed a short-term fuzzy
trading system (STFS) with technical based trading
strategy and portfolio management; its fuzzy trading
strategy uses set of appropriate technical indicators.
Wherefore, general index of the Greek (Athens)
stock market (ASE) is the validation data used by
STFS [14]. Therefore, the proposed system in this
research will be tested with the same dataset too
without consideration of the transaction cost.
Chourmouziadis [14] examined his system
separately for bull and bear market periods between
15 November 1996 and 5 June 2012 covering
different world and local economic events and
changes. The ASE demonstrated different behavior
through this period; it shows relatively long periods
of bearish market before turning to a bullish trend
then back to bearish again. It will be good test for
the systems profitability in different situations.
The three decision systems (Fuzzy RC, Fuzzy R,
and Classical) are distinguishable in their using to
the fuzzy logic. Fuzzy RC: is the first system; it is a
fuzzy decision system where fuzzy logic is applied
Received: November 30, 2017 56
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
Table 3. Systems performance: Bull markets
Bull Markets, ASE STFS Profit Classic Profit Fuzzy R Profit Fuzzy RC Profit
From To EUR % EUR % EUR % EUR %
15-11-1996 17-09-1999 63630 536.3 58167 481.67 60324 503.24 77863 678.63
01-04-2003 09-03-2009 27364 173.6 23816 138.16 25482 154.82 30144 201.44
09-03-2009 14-10-2009 15027 50.2 12812 28.12 15631 56.31 16903 69.03
Table 4. Systems performance: Bear markets
Bear Markets, ASE STFS Profit Classic Profit Fuzzy R Profit Fuzzy RC Profit
From To EUR % EUR % EUR % EUR %
20-09-1999 31-03-2003 9595 -4.05 6686 -33.14 9047 -9.53 9802 -1.98
01-11-2007 09-03-2009 7383 -26.17 7836 -21.64 6995 -30.05 7850 -21.50
15-10-2009 05-06-2012 4833 -51.67 3456 -65.44 3888 -61.12 5079 -49.21
Table 5. Proposed systems profits and drawdown
Market Periods Classic Fuzzy R Fuzzy RC
FROM TO Profit % DD % Profit % DD % Profit % DD %
15-11-1996 17-09-1999 481.67 -29.08 503.24 -25.30 678.63 -27.10
20-09-1999 20-09-1999 -33.14 -37.25 -9.53 -38.01 -1.98 -30.92
01-04-2003 01-04-2003 138.16 -24.71 154.82 -20.41 201.44 -14.06
01-11-2007 09-03-2009 -21.64 -29.52 -30.05 -26.81 -21.50 -18.32
09-03-2009 09-03-2009 28.12 -22.17 56.31 -18.24 69.03 -11.51
15-10-2009 05-06-2012 -65.44 -23.36 -61.12 -19.75 -49.21 -10.67
on the information part for both the rules and
indicators and on the estimation of the capital to
invest using Eqs. (17) and (18). Fuzzy R: is the
second system; it is also a fuzzy intelligent decision
system but fuzzy logic is used only for the
information part including the rules and technical
indicators. Although, the estimation of the capital to invest
is done using the Partial-F with value equal to 12%
of optimal-F. The third system is a classical trading
system in absence of fuzzy logic in any of its parts.
A drawdown (DD) during a specific recorded period
is a measure in percent of the decline happens in an
investment from peak-to-trough for that period.
Trading systems are run over the validation period
for the selected index (ASE), slippage adjusted and
profits are calculated on the basis of a notional
€10,000 (EUR) of the trading capital. Furthermore,
the largest drawdown is calculated. Drawdown and
trading profit for each trading period are presented
in Table 5. The Fuzzy RC showed better
performance than the Fuzzy R and Classical systems.
The total net profit achieved by the Fuzzy RC
system in bull markets and the losses in bear
markets are the best in comparison to the other
systems (Tables 3 and 4). The profit factor, the
average profit over average loss ratio, performed by
Fuzzy RC system is greater than 1, which resulted in
a profitable trading system for the whole validation
period. Figure (6) shows the performance impact of
the three trading systems and SFTS [14] on the
investment portfolio.
The final results indicate that the Fuzzy RC
produced the best performance with total net profit
€145045.26 (1450.5%). Fuzzy R ended the
validation period with total net profit €59119.25
(491.2%) which is better than classical results with
final profit €22136.3 (221.3%). The STFS [14]
performance is better than the Fuzzy R and classical
systems but less than the Fuzzy RC system and that
might be of the STFS [14] fuzzy capital
management. Therewith, the Fuzzy RC showed
better results in all trading periods than the STFS
[14] including bear and bull markets. Therefore the
importance of a trading system with those properties
resulted in high trading operation (Figs. 4 and 5).
The proposed systems are examined for the
associated maximum drawdown (DD) which is quite
high for each system. In the classical trading system
fewer trades are triggered which result in less profit
and higher DD. But effect of high DD in the Fuzzy
RC trading systems is compromised and over
challenged with much higher profitable trades. The
cumulative performance for ASE index alleviates
the DD effect when the negative trades are not
occurred in different period of profitable trades
(Table 5). The results proved that the Fuzzy RC
trading system is superior in comparison to the other
two systems. It can be considered a very good
trading system as it obtains higher profits and less
risk. Nevertheless the behavior of the Fuzzy R is
slightly better than the classic technique indicating
that the combination of both the fuzzy indicators and
Received: November 30, 2017 57
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
Figure. 4 Systems profit in bull marekts periods
Figure. 5 Systems profit in bear marekts periods
Figure. 6 Systems toal profit % for the whole validation
period
the fuzzy capital management has a significant
impact on the Fuzzy RC system performance.
With its intelligent decision, the Fuzzy RC
trading system behaves well when classical system
loses money avoiding high drawdown when the
market goes down. Fuzzy RC performance is
smooth and steady upward during the whole
validation period. These features are quite desirable.
The system produces steady profitability and
robustness under different market conditions.
More attention should be paid to the use of fuzzy
logic in trading systems not only for the indicators
and rules but also for capital management. These
experiments evidence maybe is not enough.
6. Conclusions
Fuzzy logic, by its nature, is a suitable technique
to deal with uncertainty, impreciseness and
vagueness problems; hence trading systems too. In
this research, a short-term fuzzy decision system
with fuzzy technical indicators and fuzzy trading
rules in addition to fuzzy capital management is
proposed. The system trading strategy is based on a
set of known technical indicator and new developed
one. The role of the trading system is to forecast
market moves and to give advice for issuing
positions with exit conditions. Additionally it gives
recommendations on the amount of portfolio to
invest. Therefore it is an important part of the fuzzy
decision system as most of the proposed fuzzy
decision systems in the literature don’t include fuzzy
capital management. Although, there are some
articles proposed capital management from a fuzzy
perspective [20, 22]. The validation of the trading
systems has been done for general index of the
Greek (Athens) stock market (ASE) over a period of
fifteen years. The proposed system showed
substantial performance with overall robust profit
and low drawdown. The achieved results considered
as evidence to encourage using fuzzy decision
systems with fuzzy capital management. It is a
promising techniques and merits further
investigation. Regarding this, the proposed trading
system requires more testing cases to further explore
the system properties. The future work should
include expansion of the experimental setup to cover
independent assets of different sectors over other
time periods. Furthermore it should belong with
different real portfolios.
One of the advantages of trading tools based on
fuzzy logic is the flexibility of controlling the
decisions by the operator. Wherever, future research
should concentrate on bringing this additional
flexibility. That can be done by fuzzifying other
classic trading strategies using the proposed ideas in
this research. This would give us the chance to
validate more the fitness of the proposed technique
here. Nevertheless; an optimization to the technical
indicators parameters can be done using techniques
such as genetic algorithm.
Finally, the use of fuzzy logic can be extended to
implement better risk control mechanism. That may
help in determining the exit conditions with lower
drawdown.
References
[1] R. Balvers, Y. Wu, and E. Gilliland, “Mean
reversion across national stock markets and
Received: November 30, 2017 58
International Journal of Intelligent Engineering and Systems, Vol.11, No.3, 2018 DOI: 10.22266/ijies2018.0630.06
parametric contrarian investment strategies”,
The Journal of Finance, Vol. 55, No. 2, pp.
745–772, 2000.
[2] A. Rahman and S Sasdi, "Random walk and
breaking trend in financial series: An
econometric critique of unit root tests", Review
of Financial Economics, Vol. 17, No. 3, pp.
204–212, 2008.
[3] W. Chiang, D. Enke, T. Wu, and R. Wang, "An
adaptive stock index trading decision support
system", Expert Systems with Applications, Vol.
59, No. 15, pp. 195-207, 2016.
[4] G. Armano, M. Marchesi, and A. Murru, "A
hybrid genetic-neural architecture for stock
indexes forecasting", Information Sciences, Vol.
170, pp. 3-33, 2005.
[5] P. Chang, C. Fan, and J. Lin, "Trend discovery
in financial time series data using a case based
fuzzy decision tree", Expert Systems with
Applications, Vol. 59, No. 15, pp. 195-207,
2016.
[6] A. Tealab, A. Badr, and M. Fakhr, "Automatic
FOREX Market Trading based on Technical
Analysis", Computing and Information Systems
Journal, Vol. 15, No. 1, 2012.
[7] G. Panda, G. Sahoo, and R. Majhi,
"Development and performance evaluation of
FLANN based model for forecasting of stock
markets", Expert Systems with Applications,
Vol. 36, No. 3, pp. 6800-6808, 2009.
[8] B. Malkiel, "Reflections on the Efficient
Market Hypothesis: 30 Years Later", Financial
Review, Vol. 40, pp. 1–9, 2005.
[9] C. Park and S. Irwin, "The Profitability of
Technical Trading Rules in Us Futures
Markets: A Data Snooping Free Test",
http://dx.doi.org/10.2139/ssrn.722264, 2005.
[10] M. Qiu and Y. Song, "Predicting the Direction
of Stock Market Index Movement Using an
Optimized Artificial Neural Network Model",
PLOS ONE, Vol. 11, No. 5, 2016.
[11] E. Berner, Clinical decision support systems,
2nd ed., Springer, Ed. New York, NY: Springer
Science and Business Media, LLC, 2007.
[12] X. Yong, W. Hongwei, and F. Qi, "Electronic
market models for decision support systems on
the Web", Journal of Systems Engineering and
Electronics, Vol. 15, No. 2, pp. 135-141, 2004.
[13] S. Huang, Y. Hung, and D. Yen, "A study on
decision factors in adopting an online stock
trading system by brokers in Taiwan", Decision
Support Systems, Vol. 40, No. 2, pp. 315–328,
2005.
[14] P. Chatzoglou and K. Chourmouziadis, "An
intelligent short term stock trading fuzzy
system for assisting investors in portfolio
management", Expert Systems with
Applications, Vol. 43, pp. 298-311, 2016.
[15] G. Box, G. Jenkins, G. Reinsel, and G. Ljung,
Time series analysis: Forecasting and control,
3rd ed.: Hoboken, NJ: John Wiley & Sons,
2013.
[16] L. Zadeh, "Fuzzy Sets", Information and
Control, Vol. 8, pp. 338–353, 1965.
[17] B. Nair, M. Minuvarthini, B. Sujithra, and V.
Mohandas, "Stock Market Prediction Using a
Hybrid Neuro-fuzzy System", In: Proc. of
International Conf. On Advances in Recent
Technologies in Communication and
Computing, pp. 243 – 247, 2010.
[18] L. Maciel, F. Gomide, and R. Ballini, "Stock
market volatility prediction using possibilistic
fuzzy modeling", In: Proc. of International
Conf. on Computational Intelligence, LA, 2015.
[19] S. Bharathi and A. Geetha, "Sentiment Analysis
for Effective Stock Market Prediction",
International Journal of Intelligent Engineering
and Systems, Vol. 10, No. 3, pp. 146-154, 2017.
[20] H. Kwaśnicka and M. Ciosmak, "Intelligent
Techniques in Stock Analysis", Intelligent
Information Systems, Vol. 10, pp. 195-208,
2001.
[21] R. Vince, Portfolio management formulas, New
York: Willey & Sons, 1990.
[22] C. Lin, "A fuzzy decision support system for
strategic portfolio management", Decision
Support Systems, Vol. 38, No. 3, pp. 383 - 398,
2004.